Classifications

• • G—PHYSICS
  • G01—MEASURING; TESTING
  • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
  • G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
  • G01N33/48—Biological material, e.g. blood, urine; Haemocytometers
• • G—PHYSICS
  • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
  • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
  • G16B40/00—ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding
  • G16B40/10—Signal processing, e.g. from mass spectrometry [MS] or from PCR
• • G—PHYSICS
  • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
  • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
  • G16B40/00—ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding
• • G—PHYSICS
  • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
  • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
  • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
  • G16H10/40—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for data related to laboratory analysis, e.g. patient specimen analysis
• • G—PHYSICS
  • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
  • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
  • G16B25/00—ICT specially adapted for hybridisation; ICT specially adapted for gene or protein expression
• • G—PHYSICS
  • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
  • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
  • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
  • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
• • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
  • Y10—TECHNICAL SUBJECTS COVERED BY FORMER USPC
  • Y10S—TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
  • Y10S707/00—Data processing: database and file management or data structures
  • Y10S707/99931—Database or file accessing
  • Y10S707/99933—Query processing, i.e. searching
• • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
  • Y10—TECHNICAL SUBJECTS COVERED BY FORMER USPC
  • Y10S—TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
  • Y10S707/00—Data processing: database and file management or data structures
  • Y10S707/99941—Database schema or data structure
  • Y10S707/99943—Generating database or data structure, e.g. via user interface

Definitions

• the field of the invention concerns a process for determining a biological state through the discovery and analysis of hidden or non-obvious, discriminatory biological data patterns.
• the biological data can be from health data, clinical data, or from a biological sample, (e.g., a biological sample from a human, e.g., serum, blood, saliva, plasma, nipple aspirants, synovial fluids, cerebrospinal fluids, sweat, urine, fecal matter, tears, bronchial lavage, swabbings, needle aspirantas, semen, vaginal fluids, pre-ejaculate, etc.), etc. which is analyzed to determine the biological state of the donor.
• the biological state can be a pathologic diagnosis, toxicity state, efficacy of a drug, prognosis of a disease, etc.
• the invention concerns analytical methods that a) discover hidden discriminatory biological data patterns (e.g., patterns of protein expression in a serum sample that classify the biological state of an organ) that are subsets of the larger data stream, said discrimination implying the ability to distinguish between two or more biological states in a learning set of data and b) the application of the aforementioned patterns to classify unknown or test samples. More specifically, the invention concerns a method for analysis of a data stream, which is derived from a physical or chemical analysis of molecules (e.g., proteins, peptides, DNA, RNA, etc.) in the biological sample (e.g., a mass spectrum analysis of the sample).
• hidden discriminatory biological data patterns e.g., patterns of protein expression in a serum sample that classify the biological state of an organ
• the invention concerns a method for analysis of a data stream, which is derived from a physical or chemical analysis of molecules (e.g., proteins, peptides, DNA, RNA, etc.) in the biological sample (e.
• These patterns are defined as “hidden” because they are often buried within a larger highly complex data set and are not obvious or apparent to the eye or other current classification systems.
• the pattern itself can be defined as the combination of three or more values such that the position of the vectors in an n-dimensional space is discriminatory between biological states even when individual values may not be discriminatory.
• the discriminatory patterns of the invention are novel because they can be defined without any knowledge of the identity or relationship between the individual data points in the biological data or any knowledge of the identity or relationship between the molecules in the biological samples.
• One analytical method to discover such biological states comprises the application of two related heuristic algorithms, a learning algorithm and a diagnostic algorithm, wherein the parameters of the diagnostic algorithm are set by the application of the learning algorithm to a learning set of data such that two or more biological states may be distinguished.
• biological states may be the presence or absence of a disease, efficacy or non-efficacy of a drug, toxicity or non-toxicity of a drug, etc.
• the invention is generic, specific implementations for diagnosis of various cancers (including, but not limited to carcinomas, melanomas, lymphomas, sarcomas, blastomas, leukemias, myelomas, neural tumors, etc., and cancers of organs like the ovary, prostate, and breast.), presence of a pathogen, and toxicity are disclosed.
• the preferred embodiment of the invention is the discovery and use of molecular patterns that reflect the current or future biological state of an organ or tissue.
• Another embodiment of the invention is the combination of data describing the molecular patterns of a biological state with other non-biological or clinical data (e.g., psychiatric questioning) to yield a classification describing the health of a patient.
• the detection of changes in biological states, particularly the early detection of diseases has been a central focus of the medical research and clinical community.
• the prior art includes examples of efforts to extract diagnostic information from the data streams formed by physical or chemical analysis of tissue samples. These techniques are generically termed “data-mining.”
• the data streams that have been mined are typically of two forms: analysis of the levels of mRNA expression by hybridization to DNA oligonucleotide arrays (“DNA microarrays”) and analysis of the levels of proteins present in a cell or in a serum sample, wherein the proteins are characterized either by molecular weight using mass spectroscopy or by a combination of molecular weight and charge using a 2-D gel technique.
• Liotta and Petricoin provide additional examples of protein based diagnostic methods using both 2-D gels and mass spectroscopy.
• the analysis of Liotta and Petricoin is similar to Parekh in that it consists of a search for specific tumor markers.
• Efforts to identify tumor markers have also been performed using DNA microarrays.
• Loging, W. T., 2000, Genome Res. 10, 1393-02 describes efforts to identify tumor markers by DNA microarrays in glioblastoma multiforme.
• Heldenfalk, I., et al., 2001, New England J. Med. 344, 539 report efforts to identify tumor markers that distinguish the hereditary forms of breast cancer resulting from BRCA1 and BRCA2 mutations from each other and from common idiopathic breast cancer by data-mining of DNA microarray data.
• Each vector is composed of scalars that individually correspond to the relative levels of expression of the gene under a variety of different conditions.
• Brown et al. analyzes vectors in a 79 dimension vector space where each dimension corresponds to a time point in a stage of the yeast life-cycle and each of 2,467 vectors correspond to a gene.
• the pattern recognition algorithms are used to identify clusters of genes whose expression is correlated with each other. Because the primary concern is the correlation of gene expression, the metric that is employed in the pattern recognition algorithms of Eisen et al. and related works is a Pearson coefficient or inner product type metric, not a Euclidean distance metric.
• the pattern recognition techniques of Eisen et al. were applied by Alizadeh and Staudt to the diagnosis of types of malignancy.
• Alizadeh and Staudt began by constructing vectors, each corresponding to a gene, having scalars that correspond to the relative level of expression of the gene under some differentiation condition, e.g., resting peripheral blood lymphocyte or mitogen stimulated T cells.
• the pattern recognition algorithm then clusters the genes according to the correlation of their expression and defines a pattern of expression characteristic of each differentiation state.
• Samples of diffuse large B-cell lymphomas (DLBCL) were then analyzed by hybridization of mRNA to the same DNA microarrays as used to determine the gene clusters.
• DLBCL diffuse large B-cell lymphomas
• DLBCL were found to have at least two different gene expression patterns, each characteristic of a normal differentiation state.
• the prognosis of the DLBCL was found to correlate with the characteristic differentiation state.
• the diagnostic question posed and answered in Alizadeh and Staudt was not benign or malignant but rather of determining the type or subtype of malignancy by identifying the type of differentiated cell having a pattern of gene expression most similar to that of the malignancy.
• data-mining methods based on the physical or chemical analyses having large numbers, i.e., greater than 1,000, of data points consist of two types: data-mining to identify individual markers such as genes or proteins having expression levels that are increased or suppressed in malignant cells of a defined type compared to normal cell; and data-mining wherein a pattern of known gene expression characteristic of a normal differentiated cell type is used to classify a known malignant cell according to the normal cell type it most closely resembles.
• gene expression cluster analysis is limited in scope because such analysis incorporates an analysis of all expressed genes irrespective of whether the expression of such genes is causative or merely influenced by the causative action of those genes that are characteristic of the biological state.
• the clustering analysis does not incorporate solely those genes that are characteristic of the biological state of interest, but uses the entire range of data emanating from the assay, thus making it complex and cumbersome.
• gene expression analysis must involve nucleic acid extraction methods, making it complex, and time-consuming. Pattern recognition algorithms when applied are also rendered difficult because the correlation of gene expression that is employed is a complex Pearson coefficient or inner product type metric, and not a simple Euclidean distance metric.
• the current invention discovers optimal hidden molecular patterns as subsets within a larger complex data field, whereby the pattern itself is discriminatory between biological states.
• the current invention avoids all the aforementioned problems associated with the analytical methods disclosed in the prior art, and has the ability to discover heretofore unknown diagnostic patterns.
• Such hidden molecular patterns are present in data streams derived from health data, clinical data, or biological data.
• Biological data may be derived from simple biological fluids, such as serum, blood, saliva, plasma, nipple aspirants, synovial fluids, cerebrospinal fluids, sweat, urine, fecal matter, tears, bronchial lavage, swabbings, needle aspirantas, semen, vaginal fluids, pre-ejaculate, etc., making routine sampling easy, although the expression of such molecular patterns are characteristic of disease states of remote organs. No prior knowledge of specific tumor markers or the relationship of molecules present in the biological sample to each other is required or even desired.
• the current invention also discloses methods of data generation and analysis. Such methods of data analysis incorporate optimization algorithms in which the molecular patterns are recognized, and subjected to a fitness test in which the fitness pattern that best discriminates between biological states is chosen for the analysis of the biological samples.
• the invention comprises the use of pattern discovery methods and algorithms to detect subtle, if not totally hidden, patterns in the expression of certain molecules in biological samples that are potentially diagnostic in nature, or predictive of a biological state.
• patterns of molecular expression are patterns of protein expression, particularly patterns of low molecular weight proteins (i.e. less than 20,000 Da).
• Such hidden patterns of protein expression may be obtained from only a sub-set of the total data-stream provided to the algorithm, several subsets, or may be obtained from an analysis of the total data stream.
• the pattern can be defined as a vector of three or more values such that the position of the vectors in an n-dimensional space is discriminatory between biological states even when individual values may not be discriminatory.
• the molecules of interest may be any relevant biological material such as proteins (full, cleaved, or partially expressed), peptides, phospholipids, DNA, RNA, etc.
• the discriminatory patterns that discriminate between biological states are often small subsets of data hidden in the larger data stream derived from physical or chemical analysis of the biological sample.
• a means for finding an optimal set of features that make up the discriminatory pattern is required.
• the invention incorporates the process for finding this optimal set of features.
• a number of feature selection methods for discriminatory patterns may be used to practice the invention with varying degrees of classification success. These include, but are not limited to, statistical methods, stepwise regression methods, linear optimization methods, etc.
• statistical methods have some limitations in that they are often linear, at least in their simple, well-known forms such as multivariate linear regressions.
• statistical models tend not to be robust with respect to non-linear data.
• the number of independent variables a statistical model can successfully employ is generally ten or less, with a practical preferred limit of five or six.
• the preferred embodiment uses a method that couples the genetic algorithm, an evolutionary computation method, directly to an adaptive pattern recognition algorithm to efficiently find the optimal feature set. See U.S. Patent Application titled “Heuristic Method of Classification,” (filing date: Jun. 19, 2001, claiming priority of application Ser. No. 60/212,404, filed Jun. 19, 2000).
• the diagnostic algorithm is generated by the application of the learning algorithm to a learning (or training) data set.
• the learning data set is a data set formed from biological samples for which the biological state of interest is provided for the pattern discovery operation.
• the learning data set may comprise data taken from the sera of individuals with an established biopsy diagnosis, e.g., a benign tumor and a malignant tumor. This would enable the learning algorithm to find a signature pattern of proteins that could discriminate normal from cancerous sera samples.
• the method according to the invention begins by subjecting a biological sample to a high throughput physical or chemical analysis to obtain a data stream.
• data streams include, but are not limited to, mass spectral data of proteins found in the sample or in the intensity of mRNA hybridization to an array of different test polynucleotides.
• the data stream is characterized by a large number (10,000 or more) of intensities which are generated in a way that allow for the corresponding individual datum in data streams of different samples to be identified.
• the first step of the diagnostic method is to calculate a vector, i.e., an ordered set of a small number (between 2 and 20100, more typically between 5 and 208) that is characteristic of the data stream.
• a vector i.e., an ordered set of a small number (between 2 and 20100, more typically between 5 and 208) that is characteristic of the data stream.
• Abstraction The transformation of the data steam into a vector is termed “abstraction.”
• abstraction is performed by selection of a small number of specific intensities from the data stream.
• the second step of the diagnostic method is to determine in which, if any, data cluster the vector rests.
• Data clusters are mathematical constructs that are the multidimensional equivalents of non-overlapping “spheres” of fixed size in the vector space. Such data clusters are known as hyperspheres.
• the location and associated diagnosis of each data cluster is determined by the learning algorithm from the training data set. If the vector of the biological sample lies within a known cluster, the sample is given the diagnosis associated with that cluster.
• a diagnosis can be made that the sample does not meet that classification criteria or that it is of an unspecified atypia, i.e., an “a typical sample, NOS.”
• a biological sample taken from a patient does not meet the classification of a malignant state for a specified cancer, it will be classified as non-malignant non-normal or of an unspecified atypia, “a typical sample, NOS.”
• the learning algorithm utilizes a combination of known mathematical techniques and two pre-set parameters.
• the user pre-sets the number of dimensions of the vector space and the size of the data clusters.
• the vector space is a normalized vector space such that the variation of intensities in each dimension is constant.
• the size of the cluster can be expressed as a minimum percent similarity among the vectors resting within the cluster.
• the learning algorithm contains of two generic parts, which have been developed by others and are well known in the field—a genetic algorithm (J. H. Holland, Adaptation in Natural and Artificial Systems, MIT Press 1992) and a self-organizing adaptive pattern recognition system (T. Kohonen, Self Organizing and Associative Memory, 8 Series in Information Sciences, Springer Verlag, 1984; Kohonen, T, Self-organizing Maps, Springer Verlag, Heidelberg 1997). Genetic algorithms organize and analyze complex data sets as if they were information comprised of individual elements that can be manipulated through a computer driven natural selection process.
• the search for hidden or subtle patterns of molecular expression that are, in and of themselves “diagnostic” is qualitatively different from those generated by prior art implementations of learning algorithms or data-mining techniques.
• Previous implementations of data-mining have identified specific molecular products that are indicative of a classification, e.g., proteins or transcripts that are elevated or depressed in pathological conditions.
• the level of the identified molecular products is termed per se diagnostic, because the level of the product is diagnostic without any further consideration of the level of any other molecular products in the sample, other than perhaps a normalizing molecular product that is used to normalize the level of the molecular products.
• One example of such per se diagnostic molecular products are tumor markers.
• the diagnostic significance of the level of any particular marker e.g., a protein or transcript is a function of the levels of the other elements that are used to calculate the sample vector.
• Such products are termed hereinafter as contextual diagnostic products.
• the likeness between the biological sample of interest and the learning data set was based on the specified groupings of the biological sample compared to the specified diagnostic molecular products.
• the learning algorithm discovers wholly new classification patterns without knowing any prior information about the identity or relationships of the data pattern, i.e., without prior input that a specified diagnostic molecular product is indicative of a particular classification.
• the present invention is based, in part, on the unexpectedness or non-obvious discovery of finding hidden contextual diagnostic patterns to yield a classification, e.g., the diagnosis of malignancy in cancers such as carcinomas, melanomas, lymphomas, sarcomas, blastomas, leukemias, myelomas, and neural tumors.
• cancers such as carcinomas, melanomas, lymphomas, sarcomas, blastomas, leukemias, myelomas, and neural tumors.
• the invention comprises a) creating a data stream representing the biological data (or combinations of data streams representing the biological data with clinical, health, or non-biological data) and abstraction of that data into characteristic vectors; b) the discovery of hidden diagnostic patterns of molecular expression (i.e. pattern discovery); and c) determining which biological state of interest such a pattern of molecular expression represents.
• the molecules of interest may comprise, but are not limited to, proteins, peptides, RNA, DNA, etc.
• the biological samples comprise, but are not limited to serum, blood, saliva, plasma, nipple aspirants, synovial fluids, cerebrospinal fluids, sweat, urine, fecal matter, tears, bronchial lavage, swabbings, needle aspirantas, semen, vaginal fluids, pre-ejaculate, etc.
• the biological states of interest may be a pathologic diagnosis, toxicity state, efficacy of a drug, prognosis of a disease, stage of a disease, biological state of an organ, presence of a pathogen (e.g., a virus), toxicity of one or more drugs, etc.
• the invention may be used for the diagnosis of any disease in which changes in the patterns of expression of certain molecules like proteins allow it to be distinguished from a non-diseased state.
• any disease that has a genetic component in which the genetic abnormality is expressed, one in which the expression of drug toxicity is observed, or one in which the levels of molecules in the body are affected may be studied by the current invention.
• Such diseases include, but are not limited to, cancers (carcinomas, melanomas, lymphomas (both Hodgkin's and non-Hodgkin's type), sarcomas, blastomas, leukemias, myelomas, and neural tumors, such as glioblastoma, etc.), Alzheimer's disease, arthritis, glomeruulonephritis, auto-immune diseases, etc.
• carcinomas include, but are not limited to, carcinomas of the pancreas, kidney, liver and lung; gastrointestinal carcinomas.
• the present invention is particularly valuable for the diagnosis of specific diseases for which early diagnosis is important, but is technically difficult because of the absence of symptoms, and for which the disease may be expected to produce differences that are detectable in the serum because of the metabolic activity of the pathological tissue.
• the early diagnosis of malignancies are a primary focus of the use of the invention.
• the data stream can be any reproducible physical or chemical analysis of the biological sample that results in a high throughput data stream.
• the high throughput data stream is characterized by 1,000 or more measurements that can be quantified to at least 1 part per thousand (three significant figures) and more preferably one part in 10,000.
• time of flight mass spectra of proteins may be used to generate a data stream. More specifically, matrix assisted laser desorption ionization time of flight (MALDI-TOF) and surface enhanced laser desorption ionization time of flight (SELDI-TOF) spectroscopy may be used when the molecules of interest are proteins or peptides.
• MALDI-TOF matrix assisted laser desorption ionization time of flight
• SELDI-TOF surface enhanced laser desorption ionization time of flight
• SELDI-TOF may be used to generate data streams for biological states representing toxicity, and detection of pathogens.
• data streams may be generated using serially amplified gene expression (SAGE) for gene expression classification.
• SAGE serially amplified gene expression
• data streams may be generated using 2-D Gels such as two-dimensional polyacrylamide gel electrophoreses (2D-PAGE).
• the preferred patient sample for analysis is serum.
• biopsy specimens that are relatively homogenous may also be used.
• other fluids e.g., synovial fluid may be used in the differential diagnosis of arthritis or urine in the differential diagnosis of glomerulonephritis.
• proteins that are included in either SELDI-TOF and MALDI-TOF analysis depend upon the surface or matrix that is employed. Lipophylic surfaces such as C-18 alkane surfaces are particularly convenient compared to anionic or cationic surfaces. However, those skilled in the art will appreciate that multiple spectra can be generated from the same sample using different surfaces. These spectra can be concatenated to yield “superspectra” which can be analyzed according to the invention. Likewise, data from two or more high throughput assay methods can also be joined which can be analyzed by the invention. Furthermore, biological data as described in this invention can be joined with clinical, health, or non-biological data.
• the data stream can also include measurements that are not inherently organized by a single ordered parameter such as molecular weight, but have an arbitrary order.
• DNA microarray data that simultaneously measures the expression levels of 2,000 or more genes can be used as a data stream when the tissue sample is a biopsy specimen, recognizing that the order of the individual genes in the data stream is arbitrary.
• the first step in the diagnostic process of the invention is the transformation or abstraction of the data stream into a characteristic vector.
• the data may be conveniently normalized prior to abstraction by assigning the overall peak an arbitrary value of 1.0 and, thus, all other points fractional values.
• the most simple abstraction of TOF mass spectrum consists of the selection of a small number of data points.
• more complex functions of multiple points could be constructed, such as averages over intervals or more complex sums or differences between data points that are at predetermined distance from a selected prototype data point.
• Such functions of the intensity values of the data stream could also be used and are expected to function equivalently to the simple abstract illustrated in the working examples.
• Pattern discovery is achieved by numerous methods as discussed in the Summary above.
• the pattern discovery comprises a diagnostic algorithm and a learning algorithm.
• the routine practitioner in order to practice this embodiment of the invention the routine practitioner must develop a diagnostic algorithm by employing a learning algorithm.
• the routine practitioner uses a training data set and must select two parameters, the number of dimensions and the data cluster size. See U.S. Patent Application titled “Heuristic Method of Classification,” (filing date: Jun. 19, 2001, claiming priority of application Ser. No. 60/212,404, filed Jun. 19, 2000).
• the learning algorithm can be implemented by combining two different types of publicly available generic software, which have been developed by others and are well known in the field—a genetic algorithm (J. H. Holland, Adaptation in Natural and Artificial Systems , MIT Press 1992) that processes a set of logical chromosomes 1 to identify an optimal logical chromosome that controls the abstraction of the data steam and a adaptive self-organizing pattern recognition system (see, T.
• logical chromosome is used in connection with genetic learning algorithms because the logical operations of the algorithm are analogous to reproduction, selection, recombination and mutation. There is, of course, no biological embodiment of a logical chromosome in DNA or otherwise.
• the genetic learning algorithms of the invention are purely computational devices, and should not be confused with schemes for biologically-based information processing.
• the genetic algorithm essentially determines the data points which are used to calculate the characteristic vector.
• the list of the specific points to be selected is termed a logical chromosome.
• the logical chromosome contains as many “genes” as there are dimensions of the characteristic vector. Any set of the appropriate number of data points can be a logical chromosome, provided only that no gene of a logical chromosome is duplicated. The order of the genes has no significance to the invention.
• Genetic algorithms can be used when two conditions are met.
• a particular solution to a problem must be able to be expressed by a set or string of fixed size of discrete elements, which elements can be numbers or characters, and the strings can be recombined to yield further solutions.
• One must also be able to calculate a numerical value of the relative merit of each solution, namely its fitness. Under these circumstances, the details of the genetic algorithm are unrelated to the problem whose solution is sought. Accordingly, for the present invention any generic genetic algorithm software may be employed.
• the algorithms PGAPack libraries available from Argonne National Laboratory is suitable. The calculation of the fitness of any particular logical chromosome is discussed below.
• a training data set of about 100 sample data streams was used, each sample data stream containing about 15,000 data points.
• the genetic algorithms were initialized with about 1,500 randomly chosen logical chromosomes. As the algorithm progressed, the more fit logical chromosomes are duplicated and the less fit are terminated. There is recombination between logical chromosomes and mutation, which occurs by the random replacement of an element of a logical chromosome. It is not an essential feature of the invention that the initially selected collection of logical chromosome be random. Certain prescreening of the total set of data streams to identify those data points having the highest variability may be useful, although such techniques may also introduce an unwanted initialization bias. The best fitted pattern that survives this process is used to discriminate between biological states and determine the desired classification.
• the fitness score of each of the logical chromosomes that are generated by the genetic algorithm is calculated.
• the calculation of the fitness score requires an optimal set of data clusters be generated for the given logical chromosome.
• Data clusters are simply the volumes in the vector space in which the characteristic vectors of the training data set rest.
• the method of generating the optimal set of data clusters is not critical to the invention and will be considered below. However, whatever method is used to generate the data cluster map, the map is constrained by the following rules: (i) each data cluster should be located at the centroid of the data points that lie within the data cluster; (ii) no two data clusters may overlap; and (iii) the dimension of each cluster in the normalized vector space is fixed prior to the generation of the map.
• the routine practitioner must use a learning data set and select two parameters, the number of dimensions and the data cluster size. Both parameters can be set using routine experimentation. Although there is no absolute or inherent upper limit on the number of dimensions in the vector, the learning algorithm itself inherently limits the number of dimensions in each implementation. If the number of dimensions is too low or the size of the cluster is too large, the learning algorithm fails to generate any logical chromosomes that correctly classify all samples into homogeneous clusters, and conversely if the number of dimensions can be too large. Under this circumstance, the learning algorithm generates many logical chromosomes that have the maximum possible fitness early in the learning process and, accordingly, there is only abortive selection. Similarly, when the size of the data clusters is too small, the number of clusters will be found to approach the number of samples in the training data set and, again, the routine practitioner will find that a large number of logical chromosomes will yield the maximum fitness.
• a training data set can nearly always be assigned into homogeneous data clusters.
• the value of the diagnostic algorithm generated by a learning algorithm must be tested by its ability to sort a set of data other than the training data set.
• the training data is said to be overfitted by the learning algorithm. Overfitting results when the number of dimensions is too large and/or the size of the data clusters is too small.
• the method used to define the size of the data cluster is a part of the invention.
• the cluster size is defined by the maximum of the equivalent the Euclidean distance (root sum of the squares) between any two members of the data cluster.
• a data cluster size that corresponds to a requirement of 90% similarity is suitable for the invention when the data stream is generated by SELDI-TOF mass spectroscopy data.
• 90% similarity is defined by requiring that the distance between any two members of a cluster is less than 0.1 of the maximum distance between two points in a normalized vector space.
• the vector space is normalized so that the range of each scalar of the vectors within the training data set is between 0.0 and 1.0.
• the maximal possible distance between any two vectors in the vector space is then root N, where N is the number of dimensions.
• the Euclidean diameter of each cluster is then 0.1 ⁇ root (N).
• the data stream may be converted into logarithmic form if the distribution of values within the data stream is log normal and not normally distributed.
• the fitness score for that chromosome can be calculated.
• the fitness score of the chromosome roughly corresponds to the number of vectors of the training data set that rest in clusters that are homogeneous, i.e., clusters that contain the characteristic vectors from samples having a single diagnosis. More precisely, the fitness score is calculated by assigning to each cluster a homogeneity score, which varies, for example, from 0.0 for homogeneous clusters to 0.5 for clusters that contain equal numbers of malignant and benign sample vectors.
• the fitness score of the chromosome is the average fitness score of the data clusters. Thus, a fitness score of 0.0 is the most fit.
• a preferred technique for generating for generating data clusters is using the self-organizing map algorithm as developed by Kohonen. (Kohonen, T, Self-organizing maps, Springer Verlag, Heidelberg 1997). This type of technique is variously termed a “Lead Cluster Map” (“LCM”) or an “Adaptive Feature Map” can be implemented by generic software that is publicly available. Suitable vendors and products include Model 1 from Group One Software (Greenbelt, Md.) and Adaptive Fuzzy Feature Map (American Heuristics Corp.).
• the LCM has significant advantages in that it is a) it is a non-linear modeling method; b) the number of independent variables is virtually unlimited; and c) compared to other non-linear modeling techniques, the LCM has the advantage of being adaptive. It can detect novel patterns in the data stream and track rare patterns. This is particularly important in classification of biological states, viz, mutations to viruses.
• the current invention was employed to develop a diagnostic for prostatic cancer using SELDI-TOF mass spectra (MS) of 55 patient serum samples, 30 having biopsy diagnosed prostatic cancer and prostatic serum antigen (PSA) levels greater than 4.0 ng/ml and 25 normals having PSA levels below 1 ng/ml.
• MS data was abstracted by selection of 7 molecular weight values (2092, 2367, 2582, 3080, 4819, 5439 and 18,220 Da).
• the specific molecular weights are not a critical parameter of the invention and may varying depending on the absorptive surface.
• a cluster map that assigned each vector in the training data set to a homogeneous data cluster was generated. The cluster map contained 34 clusters, 17 benign and 17 malignant.
• the diagnostic algorithm was tested using 231 samples that were excluded from the training data set. Six sets of samples from patients with various clinical and pathological diagnoses were used. The clinical and pathological description and the algorithm results were as follows: 1) 24 patients with PSA>4 ng/ml and biopsy proven cancer, 22 map to diseased data clusters, 2 map to no cluster; 2) 6 normal, all map to healthy clusters; 3) 39 with benign hypertrophy (BPH) or prostatitis and PSA ⁇ 4 ng/ml, 7 map to diseased data clusters, none to healthy data clusters and 32 to no data cluster; 4) 139 with BPH or prostatitis and PSA>4 and ⁇ 10 ng/ml, 42 map to diseased data clusters, 2 to healthy data clusters and 95 to no data cluster; 5) 19 with BPH or prostatitis and PSA>10 ng/ml, 9 map to diseased data clusters none to healthy and 10 to no data cluster.
• a sixth set of data was developed by taking pre- and post-prostatectomy samples from patients having biopsy proven carcinoma and PSA>10 ng/ml. As expected, each of the 7 pre-surgical samples was assigned to a diseased data set. However, none of the sample taken 6 weeks post surgery, at a time when the PSA levels had fallen to below 1 ng/ml, were not assignable to any data set.
• the diagnostic algorithm is able to classify a significant fraction of the samples in 3), 4) and 5) to a non-cancerous, non-normal category despite the fact that such category was not presented during training. Indeed, the fact that any samples from this group would necessarily include a substantial number with occult carcinoma carriers argues that BPH or prostatitis samples should not be included in the training data set.
• the above described methods were employed to generate a diagnostic algorithm for ovarian carcinoma again using SELDI-TOF MS analysis of patient serum.
• a training set of 100 samples was used to construct a cluster set map.
• the MS data was abstracted by selection of 5 molecular weights (531, 681, 903, 1108 and 2863 m/e).
• a cluster map consisting of 15 disease clusters and 11 healthy clusters was constructed.
• 40 were assigned to diseased data clusters, leaving 10 false negative; of the 50 samples from normals, 44 were assigned to healthy data clusters leaving 6 false positives.
• the range of values of the healthy and diseased data clusters overlapped. Indeed, for 4 of the 5 molecular weights, the range for the diseased encompassed the range for the healthy data clusters. Additionally, the diagnostic patterns being detected were not caused by tumor markers, but rather by contextual diagnostic products.
• the diagnostic algorithm was tested using a further 100 samples, which were divided into three clinical, pathological groups.
• the groups and the algorithm results were as follows: 1) 50 samples from patients with no disease, 47 were assigned to healthy data clusters and 3 to disease data clusters; 2) 32 patients with ovarian carcinoma Stages II, III or IV, all of which were assigned to diseased data clusters; and 3) 18 patients with ovarian carcinoma stage I, all of which mapped to diseased data clusters.
• the optimal proteomic pattern challenged with 100 SELDI-TOF data streams from diagnosis-blinded cases was able to accurately predict the presence of ovarian cancer in all 50 cancer specimens contained within the 100 unknown test samples (50/50, 95% confidence interval 93% to 100%). This included the correct classification of 18/18 stage I cancers (95% confidence interval 82% to 100%) while maintaining specificity for the blinded cancer-free samples (47/50, 95% confidence interval 84% to 99%, overall p ⁇ 10 ⁇ 10 by chi-squared test). These results support the hypothesis that low molecular weight proteomic patterns in sera reflect changes in the pathology of tissue within an organ at a distant site. Moreover, such patterns may be sensitive indicators of early pathological changes, since they correctly classified all 18 sera from organ-confined stage I ovarian cancer specimens.
• the training set was comprised of 56 sera, 31 from asymptomatic men with biopsy-proven prostate cancer (PSA>4 ng/ml, avg. 14.5 ng/ml), and 25 age-matched men with no evidence of prostate cancer (PSA ⁇ 1 ng/ml, avg. 0.3 ng/ml).
• PSA asymptomatic men with biopsy-proven prostate cancer
• PSA ⁇ 1 ng/ml avg. 0.3 ng/ml
• the 56 sera were analyzed by SELDI-TOF.
• the pattern discovery analysis found a signature pattern of the combined normalized intensities of 7 protein peaks (out of 15,000 7 possible permutations) at the specific molecular weights of 2092, 2367, 2582, 3080, 4819, 5439, and 18220 Da that could distinguish all 56 samples analyzed in the prostate sera training set.
• the optimal proteomic pattern was tested with 227 blinded sera samples.
• the blinded study set contained a) 24 sera from asymptomatic men who had subsequent biopsy-proven cancer, and whose PSA values were between 4-10 ng/ml at the time of collection, b) control sera from 6 age-matched males (PSA ⁇ 1 ng/ml) and c) 197 sera from men with biopsy-proven benign prostatic hypertrophy or prostatitis (PSA values ranged from 0.4 ng/ml to 36 ng/ml).
• the data-mining tool was able to accurately predict the presence of prostate cancer in the blinded test set (92%, 22/24, p ⁇ 0.000001 compared to patients with BPH), including 17/18 containing PSA values of 4-10 ng/ml.
• 70% of the patients (137/197) with biopsy proven BPH were classified as belonging to a unique (non-normal, non-cancer) phenotype.
• Only 1% of the sera from the BPH-positive cohort was categorized as a normal phenotype.
• sera from 6 healthy controls were compared to those of the 24 patients with biopsy proven cancer, 6/6 healthy patients were classified correctly, compared to 22/24 patients with prostate cancer (p ⁇ 0.000001).
• the anonymized ovarian screening serum study set was obtained from the Early Detection Research Network (“EDRN”) National Ovarian Cancer Early Detection Program according to full Institutional Review Board (“IRB”) oversight.
• EDRN Early Detection Research Network
• IRB Institutional Review Board
• This set contained sera from 200 asymptomatic women, 100 with ovarian cancer at the time of sample collection and 100 control women at risk for ovarian cancer as defined by family history or previous breast cancer diagnosis (Table 3). This group of unaffected women had been followed and was disease-free for at least five years. All sera were obtained prior to diagnosis and intervention.
• the disease cohort included histology confirmed papillary serous, endometrioid, clear cell, mucinous, adenocarcinoma, and mixed ovarian cancers of all stages. All women in the disease cohort underwent extensive surgical exploration and formal FIGO staging.
• the anonymized prostate screening serum study set was obtained from a prostate cancer-screening clinic where samples were obtained under approved informed consent (277 samples) (Table 3). An additional 20 anonymized specimens were collected at the National Cancer Institute under IRB approved informed consent. The Chilean trial was initiated in 1996 and lasted for 5 years. The subject eligibility criteria required asymptomatic men over the age of 50 with no previous history of prostate cancer. All men provided a serum sample and then received a medical evaluation and a digital rectal examination. Subsequently, men with a serum PSA>4.0 ng/ml, or suspicious digital rectal examinations were subjected to a single core needle biopsy for pathologic diagnosis. The prostate adenocarcinomas represented were of a full spectrum of grades (I-III) and Gleason scores (4-9).
• the 20 sera acquired at the NCI were taken from a) 7 men at the time of diagnosis and six weeks after prostatectomies for biopsy-proven organ confined prostate cancer and b) 6 normal healthy male volunteers, PSA ⁇ 1.0 ng/ml. All sera were obtained prior to medical examination, diagnosis, and treatment. All sera were collected, spun down, aliquoted and stored in liquid nitrogen until use. Received sera were thawed once, separated into 10 microliter aliquots, and then refrozen in liquid nitrogen until SELDI-TOF analysis was performed.
• Sample preparation One microliter of acetonitrile (Sigma-Aldrich Co., St. Louis, Mo.) was added to the sample spots of the 8-feature C18 hydrophobic interaction protein chip (Ciphergen Biosystems, Inc., Freemont, Calif.). This chip will bind proteins through hydrophobic interactions that are dependent upon the intrinsic primary amino acid sequences specific for every protein.
• the acetonitrile application was followed by the addition of 1 ⁇ l of serum. The sample was allowed to air dry on the chip. The chips were vigorously washed by vortexing in deionized water for 4 minutes and allowed to air dry. Lastly, 0.5 ⁇ l of CHCA solution was added.
• SELDI-TOF like other time-of-flight spectrometric techniques, has its best sensitivity at the low molecular weight range ( ⁇ 20,000 Da). Data were recorded and optimized for analysis with the SELDI Protein Biology System version 2.0 software (Ciphergen Biosystems, Inc., Palo Alto, Calif.). Raw SELDI data, not filtered or scaled in any way, were converted to ASCII data files for analysis by the data-mining tool.
• the method of the invention was tested on data streams obtained from biological samples from rats treated with doxorubicin that developed proven cardiotoxicity. Controls were treated with saline. The biological samples obtained from rats showing cardiotoxicity were classified correctly with 100% selectivity and 100% sensitivity and no false positives. See Table 4.
• Rats were treated with doxorubicin and a cardioprotectant. Thus, some animals had toxicity while others did not. Table 8 shows that using the method of the invention all but one of the treated animals could be correctly identified, while only misclassifying 2 control animals. See Table 5.
• Nipple aspirants taken from breast cancer patients were analyzed using the process of the invention.
• the nipple aspirants were subjected to a mass spectral analysis and subjected to a pattern finding method. A sensitivity of nearly 92% was observed. See Table 8.

Landscapes

• Health & Medical Sciences (AREA)
• Engineering & Computer Science (AREA)
• Life Sciences & Earth Sciences (AREA)
• Physics & Mathematics (AREA)
• Medical Informatics (AREA)
• General Health & Medical Sciences (AREA)
• Spectroscopy & Molecular Physics (AREA)
• Epidemiology (AREA)
• Public Health (AREA)
• Biotechnology (AREA)
• Artificial Intelligence (AREA)
• Data Mining & Analysis (AREA)
• Databases & Information Systems (AREA)
• Biophysics (AREA)
• Evolutionary Computation (AREA)
• Bioethics (AREA)
• Computer Vision & Pattern Recognition (AREA)
• Software Systems (AREA)
• Bioinformatics & Cheminformatics (AREA)
• Bioinformatics & Computational Biology (AREA)
• Theoretical Computer Science (AREA)
• Evolutionary Biology (AREA)
• Molecular Biology (AREA)
• Signal Processing (AREA)
• Primary Health Care (AREA)
• Chemical & Material Sciences (AREA)
• Analytical Chemistry (AREA)
• General Physics & Mathematics (AREA)
• Hematology (AREA)
• Food Science & Technology (AREA)
• Medicinal Chemistry (AREA)
• Biomedical Technology (AREA)
• Biochemistry (AREA)
• Urology & Nephrology (AREA)
• Immunology (AREA)
• Pathology (AREA)
• Investigating Or Analysing Biological Materials (AREA)
• Measuring Or Testing Involving Enzymes Or Micro-Organisms (AREA)
• Other Investigation Or Analysis Of Materials By Electrical Means (AREA)
• Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)

Priority Applications (1)

Applications Claiming Priority (4)

Related Child Applications (1)

Publications (3)

Family

ID=30119299

Family Applications (2)

Family Applications After (1)

Country Status (15)

Cited By (80)

Families Citing this family (104)

Citations (40)

Family Cites Families (26)

• 2001
  • 2001-07-18 BR BRPI0112667-9A patent/BR0112667A/pt not_active IP Right Cessation
  • 2001-07-18 IL IL15385601A patent/IL153856A0/xx unknown
  • 2001-07-18 AU AU2001280581A patent/AU2001280581A1/en not_active Abandoned
  • 2001-07-18 EA EA200300161A patent/EA200300161A1/ru unknown
  • 2001-07-18 WO PCT/US2001/022447 patent/WO2002006829A2/en active Application Filing
  • 2001-07-18 EP EP01958977A patent/EP1386275A2/en not_active Withdrawn
  • 2001-07-18 MX MXPA03000506A patent/MXPA03000506A/es not_active Application Discontinuation
  • 2001-07-18 NZ NZ524171A patent/NZ524171A/en unknown
  • 2001-07-18 JP JP2002512687A patent/JP5246984B2/ja not_active Expired - Fee Related
  • 2001-07-18 CA CA002415775A patent/CA2415775A1/en not_active Abandoned
  • 2001-07-18 CN CNA018157548A patent/CN1484806A/zh active Pending
  • 2001-07-18 SG SG200501160-6A patent/SG144731A1/en unknown
  • 2001-07-18 US US09/906,661 patent/US6925389B2/en not_active Expired - Lifetime
  • 2001-07-18 KR KR1020037000761A patent/KR101054732B1/ko not_active IP Right Cessation
• 2003
  • 2003-01-17 NO NO20030251A patent/NO20030251L/no not_active Application Discontinuation
• 2005
  • 2005-07-27 US US11/189,808 patent/US20050260671A1/en not_active Abandoned
• 2010
  • 2010-11-23 AU AU2010246364A patent/AU2010246364A1/en not_active Abandoned
• 2012
  • 2012-12-27 JP JP2012286014A patent/JP2013101130A/ja active Pending

Patent Citations (44)

Non-Patent Citations (97)